How many Claude models are there?

Generations: Claude 1 (2023), Claude 2 (2023), Claude 2.1 (2023), Claude 3 (2024), Claude 3.5 (2024-25), Claude 4 family (2025-26). Within each generation, three size tiers: Haiku (small), Sonnet (medium), Opus (large).

What is the current flagship Claude model?

Claude Opus 4.6, released in 2026. It is Anthropic's largest, most capable model, particularly strong on long-form writing, coding, and deep reasoning.

What is the difference between Haiku, Sonnet, and Opus?

Haiku is smallest and fastest — used for low-latency API calls. Sonnet is mid-tier, strong general-purpose, fast. Opus is largest — strongest on hard problems but slower and more compute-intensive.

Which Claude model does Claude Pro use?

Claude Pro ($20) defaults to Sonnet and gives access to Opus 4.6 within generous usage limits. Max 5x ($100) raises the limits 5x; Max 20x ($200) raises them 20x and is aimed at Claude Code power users.

Does Claude have a reasoning mode?

Yes — Extended Thinking, introduced with Claude 3.7. It lets the model take more time to deliberate before answering. Claude Opus 4.6 uses Extended Thinking automatically on hard problems.

Has Anthropic deprecated any Claude models?

Yes — Claude 1 and Claude 2 were retired in 2024. Claude 3 non-Sonnet variants were deprecated through 2025. Anthropic generally gives 6-12 months notice before deprecations.

Claude Model Evolution: Sonnet, Opus 4.6

The Anthropic lineage

Every Claude model follows the same Haiku/Sonnet/Opus naming pattern (small/medium/large) and every generation has gotten noticeably better at writing, reasoning, and coding. This is the full history and what each release means for a Claude subscription.

Chronological releases

Claude 1 — March 2023

Anthropic’s first public model. Roughly comparable to GPT-3.5. Available via API and a small beta chat.

Claude 2 — July 2023

First broadly available Claude. 100K-token context window made a real splash — it could read an entire book.

Claude 2.1 — November 2023

200K-token context. Lower hallucination rates. Set Claude’s reputation for long-document work.

Claude 3 — March 2024

Three-tier launch: Haiku, Sonnet, Opus. Opus 3 briefly took the top benchmark spot from GPT-4. Ignited serious competition in the enterprise market.

Claude 3.5 Sonnet — June 2024

The most important Claude release commercially. Matched or beat Opus 3 on most tasks at Sonnet cost. Became the default model in Claude.ai and a favorite of developers.

Claude 3.5 Haiku — November 2024

Small-tier upgrade; fast, cheap, and good enough to serve a lot of the API traffic.

Claude 3.7 Sonnet — early 2025

Introduced Extended Thinking — Claude’s first reasoning mode.

Claude 4 family — 2025

Generational jump. Claude 4 Sonnet replaces 3.5 Sonnet as default. Claude 4 Opus becomes flagship.

Claude Code CLI — 2025

Not a model, but a packaging: Anthropic ships an official terminal-based agent that runs on Opus. Drives adoption of the Max 20x subscription.

Claude Opus 4.6 — 2026

Current flagship. Incremental but material upgrade on reasoning, coding, and writing quality. Available across Pro and Max tiers.

Which Claude tier powers which model

Tier	Price	Models included
Free	$0	Sonnet (limited)
Pro	$20	Sonnet, Opus 4.6 (with daily caps), Haiku
Max 5x	$100	Same models, 5x the usage
Max 20x	$200	Same models, 20x the usage — aimed at Claude Code
Team	$30/user	Sonnet + Opus with admin features
Enterprise	Custom	Full lineup + compliance + dedicated capacity

Deprecation patterns

Claude 1, Claude 2, and Claude 2.1 retired by end of 2024.
Claude 3 Opus and Claude 3 Haiku retired through 2025 as Claude 4 shipped.
Anthropic’s published deprecation policy is 6 months minimum notice for API models.
Claude.ai users are typically moved to the new default silently when a generation ships.

Why this matters for picking a subscription

If you write for a living or code seriously, you want Opus 4.6. That requires at minimum Claude Pro at $20. If you run Claude Code all day and hit Pro’s Opus caps, Claude Max 20x at $200 is the step up — it’s not cheap, but it’s the only subscription on the market aimed squarely at terminal agent workflows. Everybody else is fine on Pro.